Skip to main content

Own product

LotStudio AI

B2B SaaS for car dealerships that cleans up vehicle photos, removing or replacing backgrounds for studio-quality listings while preserving the real details of the car.

Role
Product design and build
Timeline
2026
Tools
Supabase, Render
Type
Own product
Open the live product(opens in a new tab)
LotStudio: a white sedan photographed on a dealership lot, split into the original photo and a clean studio result, with the steps take a photo, AI removes background, studio ready

The problem

Vehicle photos taken on a dealership lot come with busy backgrounds, glare and small distractions. Cleaning them up for a listing is slow by hand, and generic AI tools often change the car itself, which a listing cannot allow.

My role and constraints

LotStudio AI began as an internal tool built around a dealership’s real inventory, photos and staff workflow. It is now being turned into a SaaS product with support from management and ownership.

I designed and built it. The hard constraint: the output has to look studio quality while every real detail of the vehicle stays true.

Process

The MVP shipped on May 19, 2026: upload a lot photo, choose a background, get a cleaner listing image. About 18 commits of iteration followed, each driven by what broke in real use.

One finding shaped the pipeline: higher-quality output settings blurred vehicle details, so quality was tuned for fidelity to the car rather than for the highest setting.

Building with AI is about testing, noticing what breaks and why, and making product decisions from there. When the model redraws a dashboard or gear shifter instead of removing one distraction, that is a product architecture problem. That is where masked editing, quality control and workflow design matter.

How a photo moves through LotStudio

  1. Stage 1

    • Staff upload. Signed in with magic link or email
    • Inventory website. Sends images with an API key

  2. Stage 2

    • Photo type. Exterior, interior, detail, white interior

  3. Stage 3

    • Background. Remove or replace

  4. Stage 4

    • AI processing. Quality tuned for vehicle detail

  5. Stage 5

    • Magic Eraser. Brush a mask; only that area changes

  6. Stage 6

    • Job summary. Thumbnails and clear labels
Photos arrive from staff or straight from an inventory website, and only masked areas are ever edited by hand.

What broke, and the product decision it led to

  • What broke

    The model redrew details like a dashboard or gear shifter instead of removing one distraction.

    What I changed

    Magic Eraser masked editing: brush over the sticker, dealer card, fingerprint or glare, and only that area changes.

  • What broke

    Higher-quality output settings blurred vehicle details.

    What I changed

    Tuned output quality for fidelity to the real car instead of the highest setting.

  • What broke

    Build issues, path aliases and dependencies stood between the MVP and production.

    What I changed

    Fixed the build and deployment setup so it runs in production on Render.

The solution

  • Multiple photo types: exterior, interior, detail, and an interior white-background mode.
  • Magic Eraser masked editing: brush over a sticker, dealer card, fingerprint, glare or object so only that area changes.
  • Supabase auth with magic link and email and password sign-in.
  • API-key integration so an inventory website can send vehicle images straight in.
  • Thumbnails, processing animations, clearer job summaries and labels.
  • Deployed on Render: I fixed build issues, path aliases, dependencies and the deployment setup to make it production ready.

Outcome

LotStudio AI is live and in production on Render.

MVP shipped
May 19, 2026
Commits of iteration since MVP
~18
Photo types supported
4

What I learned

Prompting is not the product. The product is the set of guardrails around the model: what it is allowed to touch, how quality is checked, and how staff fix the edge cases.